Bug 39256 – Make sure globalExec always has a correct scopechain

• Component: New Bugs
• Status: RESOLVED INVALID

[Oliver Hunt]

Make sure globalExec always has a correct scopechain

[Oliver Hunt]

Created attachment 56310 [details]
Patch

[Darin Adler]

Comment on attachment 56310 [details]
Patch

>  ExecState* JSGlobalObject::globalExec()
>  {
> -    return CallFrame::create(d()->globalCallFrame + RegisterFile::CallFrameHeaderSize);
> +    ExecState* result = CallFrame::create(d()->globalCallFrame + RegisterFile::CallFrameHeaderSize);
> +    ASSERT(d()->globalScopeChain.node());
> +    result->init(0, 0, d()->globalScopeChain.node(), CallFrame::noCaller(), 0, 0, 0);
> +    return result;
>  }

This seems to add quite a bit of extra work to a function that used to just compute and return a pointer. Why is this better?

[Oliver Hunt]

(In reply to comment #2)
> (From update of attachment 56310 [details])
> >  ExecState* JSGlobalObject::globalExec()
> >  {
> > -    return CallFrame::create(d()->globalCallFrame + RegisterFile::CallFrameHeaderSize);
> > +    ExecState* result = CallFrame::create(d()->globalCallFrame + RegisterFile::CallFrameHeaderSize);
> > +    ASSERT(d()->globalScopeChain.node());
> > +    result->init(0, 0, d()->globalScopeChain.node(), CallFrame::noCaller(), 0, 0, 0);
> > +    return result;
> >  }
> 
> This seems to add quite a bit of extra work to a function that used to just compute and return a pointer. Why is this better?

it's groundwork for future optimisations

[Geoffrey Garen]

Can you arrange to initialize d()->globalCallFrame once?

As it stands, globalExec() will neurotically write the same ScopeChainNode* and CallFrame::noCaller() value to d()->globalCallFrame every time it's called.

[Oliver Hunt]

Comment on attachment 56310 [details]
Patch

this isn't necessary, it turns out everything is currently as we want it to be.